If a thinly traded stock has not traded for the last few days (volume=0), is it better to use the last known trade price (i.e. roll over last non-missing trade price) or use last known bid/ask/valuation price?
This is for stock chart application.
I don't trust either.
That a stock didn't trade carries information about its liquidity and about the magnitude of innovations in its fundamental value. If it is feasible within your model, try to incorporate the framework of Rosett (1959, “A Statistical Model of Friction in Economics”, Econometrica). For a recent application of the friction model to financial data, see “Corporate Yield Spreads and Bond Liquidity” (JF, by Chen, L., D. A. Lesmond, and J. Wei).
The basic idea behind using a friction model of liquidity is the following: while true returns are determined by many stochastic factors, observed returns mirror changes in the true value only if the liquidity costs are less then the profit to the marginal trader.
By stock chart application you mean you are making a charting tool for traders? Typically there is a choice to plot trade, bid or ask, and almost all the time they will want to look at trade prices. If a stock hasn't been trading then the flatline (or gap) on the chart communicates that.